Modal Analysis by Free Vibration Response Only for Discrete and Continuous Systems

Bor-Tsuen Wang 1 , Deng-Kai Cheng 2

1
Professor 2 Graduate student
Department of Mechanical Engineering
National Pingtung University of Science and Technology
Pingtung, 91201
Taiwan
TEL：886-8-770-3202 ext. 7017
FAX：886-8-7740-0142
E-mail：wangbt@mail.npust.edu.tw

Abstract:
This work aims to develop the algorithm for modal analysis by free vibration response only
(MAFVRO), in particular for the general or non-proportional viscous damping system model. If the
structural displacement or acceleration response due to free vibration can be measured, the system
response matrices, including the displacement, velocity and acceleration, can be obtained through
numerical differential or integration methods. These response matrices can then be applied to the
developed MAFVRO method to determine the structural modal parameters. The numerical
differential and integration methods are introduced and adopted to establish the modal parameter
prediction program for the non-proportional damping model of MAFVRO. This work also shows
the applications of MAFVRO to the multiple degree-of-freedom (MDOF) systems and the
cantilever beam, respectively. Both the discrete and continuous systems are demonstrated for the
feasibility of the MAFVRO algorithm. The developed method uses the free vibration output
response only and can obtain the structural modal parameters successfully.

Keywords: modal analysis; modal parameter; free vibration; non-proportional viscous damping;

1
1. Introduction

The knowledge of structural modal parameters is of great interest. Natural frequencies generally
account for the structural mass and stiffness distributions. Mode shape patterns corresponding to the
natural frequencies can be informative for structural design consideration or other monitoring
purposes. Structural modal damping ratios are also of importance for theoretical simulations, since
they are not theoretically available in general. In particular, experimentally extracted modal
parameters can be useful for analytical model validation, structural design modification, response
simulation for optimum design, force prediction, structural monitoring or damage detection, etc.
Experimental methods to identify the structural modal parameters are necessary.

In conventional experimental modal analysis (EMA) [1,2], the test structure is usually assumed
to be at rest and without other external excitation during modal testing. For practical applications
the test subject can be in operational conditions. The operational modal analysis (OMA) approach
has drawn much attention. The operational conditions can be featured in several regimes, such as
ambient excitation, natural source, free response and unknown or uncontrolled inputs. Therefore,
output-only modal analysis (OOMA) is also termed and interested.

The selection of excitation forms and sensors is of importance in EMA. Wang [3] presented the
theoretical formulation of generic frequency response functions (FRFs) for continuous systems and
provided the theoretical base for applying various forms of actuators and sensors to structural modal
testing. In EMA, the excitation should be controllable and measurable. The impact hammer is
frequently used as the actuator and the accelerometer as the sensor via a FFT analyzer to obtain the
system FRFs between the excitation and the structural response. One can obtain structural modal
parameters including natural frequencies, mode shapes and modal damping ratios using a
curve-fitting process or modal parameter extraction methods. This traditional approach is
categorized as input-output modal analysis. This work will focus on the output-only modal analysis
for free vibration response only.

For the traditional EMA requiring FRFs, the operational modal analysis for frequency domain
decomposition (FDD) methods [4], [5], that have been developed and widely adopted for modal
identification, starts from spectral density functions of system response. Cauberghe et al. [6]
presented a combined experimental-operational modal analysis method to estimate the modal
parameters by using the Fourier spectra of the outputs and known input. Magalhaes et al. [7] used
the ambient vibration test (AVT) data for the modal identification of a cable-stayed bridge by
Enhanced frequency domain decomposition (EFDD) and Stochastic Subspace Identification (SSI)
methods. Gentile and Gallino [8] also applied the peak picking method and the EFDD techniques to
extract the modal parameters from ambient vibration data for a suspension footbridge. For the
above mentioned approaches, the fast Fourier transform (FFT) operation is generally required and
causes the complexity for numerical operation of measured data in signal processing. This work

2
mainly adopts the time domain method for requiring only free vibration response data and related
matrix operations to determine structural modal parameters.

Another category of time domain modal identification approach is the autoregressive (AR)
method. Moore et al. [9] presented an autoregressive moving average with exogenous excitation
(ARMAX) time-domain parameter estimation algorithm to identify modal parameters of structures
in the presence of significant measurement noise and unmeasured sources of periodic and random
excitation. Larbi and Lardies [10] presented a time-domain procedure using the vector
autoregressive moving-average (VARMA) process for identification of modal parameters of
vibrating systems from multiple output data only based on the estimation of multivariate
autoregressive (AR) coefficients using a maximum likelihood technique. Papakos and Fassois [11]
presented a comprehensive autoregressive (AR) and linear multistage autoregressive moving
average (LMS-ARMA) method for multi-channel identification of structures under unobservable
excitation. The approach was demonstrated for modal identification of an aircraft skeleton structure.

Other special techniques in EMA are also developed. Devriendt and Guillaume [12] introduced
an approach to identify modal parameters from output-only transmissibility measurements and
demonstrated by means of an experimental test on a clamped beam. Devriendt and Guillaume [13]
developed a transmissibility-based approach for output-only modal analysis where the unknown
operational forces can be arbitrary. Parloo et al. [14] provided the estimation of the operational
scaling factor by performing the experiment adding a controlled mass on the test structure so as to
determine the normalized mode shape vector. Parloo et al. [15] further applied the idea to identify
the force in operational condition during output-only modal analysis. Abdelghani et al. [16]
presented the subspace-based damage detection and isolation algorithms for on-line structural
monitoring of an airplane structure under unknown excitation. The modal results from the
output-only subspace-based modal analysis are compared with those from the modal appropriation
method [17], which is based on a pure sinusoidal excitation of the structure to obtain the modal
parameters of one mode in each experiment.

For free vibration response modal analysis methods, Huang and Su [18] proposed an approach
based on the continuous wavelet transform (CWT) to identify modal parameters of a linear system
from its seismic response with the knowledge of source excitation and also applicable to free decay
response. Their approach employed the time invariance property and filtering ability of CWT to
improve data processing efficiency. The approach was validated to obtain modal parameters for a
three-story non-symmetric steel frame in a shaking table. Lardies [19] presented a time domain
method from multi-output sensors only for modal parameter identification. The method used
Cayley-Hamilton theorem to find a set of scalar parameters related to the modal properties of a
vibrating system and required only knowledge of the covariance matrix of signals.

Wang and Cheng [20] proposed an algorithm of modal analysis from free vibration response

3
only (MAFVRO). Their formulation is limited to the proportional viscous damping based on normal
mode analysis. The natural frequencies and mode shape vectors for MDOF systems can be
successfully obtained. This work extends the MAFVRO to the general or non-proportional viscous
damping cases. The complex mode analysis is adopted and thus the natural frequencies, complex
mode shapes and modal damping ratios can be determined, simultaneously.

The MAFVRO method is inspired from Zhou and Chelidze [21] who used the free vibration
response data and adopted smooth orthogonal decomposition (SOD) method to extract normal
modes of discrete system requiring the prior knowledge of mass matrix. Feeny and Kappagantu [22]
may be the first to discuss the modal analysis for structure in free vibration condition. They adopted
proper orthogonal decomposition (POD) or so called Karhunen Loeve decomposition (KLD) to
obtain mode shape vectors only. Han and Feeny [23] applied POD to obtain proper orthogonal
mode (POM) that is the structural normal mode. The practical limitation is that the mass matrix
must be proportional to the identity matrix. The MAFVRO algorithms presented in this work
require no prior knowledge of system matrices in obtaining modal parameters, in particular the
complex mode is also considered for the non-proportional damping model rather than the normal
mode analysis in the proportional damping model [20].

This paper derives the complete MAFVRO formulation for both the proportional and
non-proportional viscous damping models in Section 2 and considers both the displacement sensors
and accelerometers applications, respectively, in Section 3. Section 4 introduces the development of
MAFVRO application program in MATLAB software. Section 5 demonstrates the applications of
the MAFVRO algorithm to the MDOF discrete systems and the beam structure, i.e. a continuous
system.

2. MAFVRO formulation

Consider a MDOF vibration system with viscous damping. The general form of equation of
motion can be expressed as follows:

Mx + Cx + Kx = f . (1)

The initial conditions are

x(0) = x 0 . (2)

x (0) = v 0 . (3)

M, C and K are the n×n mass, damping and stiffness matrices of the MDOF system, respectively,
and n is the number of DOFs for the system. x 0 and v 0 are the initial displacement and velocity
vectors, respectively.

4
2.1. Proportional viscous damping model

For the proportional viscous damping, the following relation holds:

C = αM + β K (4)

where α and β are some constants. For normal mode analysis, let

x = Xe iωt . (5)
By the substitution of Eq. (5) into Eq. (1) and the assumptions of f = 0 and C = 0 , the
generalized eigenvalues problem can be formulated:

KX = ω 2 MX (6)
or

M −1 KX = ω 2 X (7)
By solving the above equation, n-pair of eigenvalues ω r2 and eigenvector X r can be obtained.

Physically, ω r = 2πf r is the r-th natural frequency in rad/sec, f r is in Hz, and X r = φ r is its

corresponding mode shape vector.

The following derivation is partly adopted from Wang and Cheng [20]. They showed the
formulation to determine modal parameters from the free vibration response, i.e. f = 0 . For the
proportional viscous damping model without the prescribed force, the system equation becomes
Mx + (αM + βK)x + Kx = 0 . (8)

Rearrange the above equation

M (x + αx ) = −K (x + βx ) . (9)

Then
M −1K = −(x + αx )(x + βx ) −1 . (10)

By comparing Equations (10) and (7), one can conclude that if the system responses x , x
and x are known, M −1 K can be formulated and used to solve for the eigenvalues and
eigenvectors. The drawback of the formulation is the requirement of prior knowledge of constants
α and β . For the proportional viscous damping model, i.e. the normal modes of the system, the

natural frequencies ωˆ r = 2πfˆr and mode shape vectors φ̂ r can be obtained, in particular φ̂ r is

real. The symbol ˆ denotes the solutions from the proportional model.
Consider the system displacement response matrix as follows:

5
 T
 x1,k x2,k  xn ,k    x1,k   x1,k +1   x1,k + Nk −1 
 x      
x2,k +1  xn ,k +1   x2,k   x2,k +1   x2,k + Nk −1 
X ]N k × n 
1, k +1
=
X(t ) [ = =          
     
        . (11)
 x1,k + Nk −1 x2,k + Nk −1  xn ,k + Nk −1    xn ,k   xn ,k +1   xn ,k + Nk −1 
    
 
T
= {x}k {x}k +1 {x}k + Nk −1 

where x r ,k = x r (t k ) denotes the displacement of the r-th DOF at time tk as depicted in Fig. 1. k

is the starting point, and N k is the total number of points adopted for MAFVRO algorithm. [ ] and
{ } denote the matrix and vector, respectively. The superscript T denotes the transpose operation on
the matrix. Similarly, the system velocity and acceleration response matrix can also be defined

 = [ X ] = [{x} {x} {x} T

X k k +1 k + N k −1 ] . (12)

 = [ X ] = [{x} {x} {x} T

X k k +1 k + N k −1 ] . (13)

Eq. (10) can then be rewritten as follows:

 T + αX
M −1K = −( X  T )(βX
 T + X T ) −1 . (14)

The measurement and derivation of those response matrices will be discussed in Section 3.

2.2. Non-proportional viscous damping model

For the general or non-proportional viscous damping, the following equilibrium equation is
invoked:

Mx − Mx = 0 . (15)

By combining Eqs. (1) and (15), the system equation can be rewritten as follows:
Ay + By = P . (16)

where
 0 M − M 0  x  0
A=  ,B =   , y =  , P =   . (17)
M C   0 K x  f 
Let
y = Ye λt . (18)

By the substitution of Eq. (18) into Eq. (16) and the assumption of zero external force vectors f =0,
i.e. P =0, the eigenvalue problem can be formulated as follows:

BY = − λAY . (19)

6
or
(− A −1 B)Y = λY . (20)

By solving the above equation, 2n-pair of complex conjugate eigenvalues and their corresponding
eigenvectors can be obtained:

 λr → Yr
 * , r = 1,2,..., n . (21)
 λr → Yr
*

where
λr
= Re ± iI m = −ζ r ω r ± i ω r 1 − ζ r2 . (22)
λr
*

The equivalent natural frequency and modal damping ratio can be determined:

ω r = Re2 + I m2 . (23)

− Re
ζr = . (24)
Re2 + I m2

and

X  X * 
Yr =  r , Yr* =  *r  . (25)
X r  X r 

where ω r = 2πf r , ζ r and φ r = X r are the r-th natural frequency, modal damping ratio and

displacement mode shape vector, respectively. The bar symbol is to denote the solutions from the
complex mode analysis, in particular for the non-proportional viscous damping.

Similar to the derivation of the proportional viscous damping model for MAFVRO, this work
is mainly to extend the MAFVRO for the non-proportional viscous damping model. The system
equation in Eq. (16) without the prescribed force, i.e. P =0, is as follows:
Ay + By = 0 . (26)

One can obtain

−1
xx 
− A B = y y −1 =    .
−1
(27)
x x 

From the definition of the system displacement, velocity and acceleration response matrices as
shown in Eqs. (11)-(13), the above equation becomes

7
 −1
X T   X
 T
−1
−A B =
  T T . (28)
X  X 

By comparing Eqs. (20) and (28), one can see that if the system response matrices are known,
−1
− A B can be formulated and used to solve the eigenvalues and eigenvectors. Therefore, the
system modal parameters as shown in Eqs. (23)-(25) can be obtained. This approach is the main
idea of MAFVRO for the non-proportional viscous damping.
Note that the MAFVRO algorithms for both proportional and non-proportional viscous
damping MDOF systems are formulated. Only the the r-th natural frequency ( ω̂ r ) and its

corresponding mode shape ( φ̂ r ) can be obtained for the proportional viscous damping model and

the normal mode analysis is assumed, i.e. φ̂ r is the real mode. For the non-proportional viscous

damping model, the complex mode analysis is adopted, so the equivalent natural frequency ( ωr )

and modal damping ratio ( ζ r ) can be obtained from Eqs. (23) and (24). The mode shape vector ( φ r )

is essentially complex. The novelty of MAFVRO for the non-proportional viscous damping in this
work is that only the system transient response, such as the displacement or acceleration, is required
to formulate the response matrices as described in Eqs. (11)-(13) and to obtain − A −1B as shown in
Eq. (28). The modal parameters of the system can be predicted as discussed. Section 3 will show the
formulation of system response matrices for different types of sensors.

It is also noted that the developed MAFVRO approach for MDOF systems can be applied to a
continuous structure as well. If the structure is divided into m measurement points, and single
axial measurement is applied at each point, then the number of measurement points m can be
considered as the DOFs of the equivalent lumped mass system, i.e. the number of DOFs becomes
n = m . The formulation of MAFVRO can still be valid for continuous systems. The case study for
the beam structure in Section 5 will be shown to demonstrate the application of MAFVRO to the
continuous system.

3. Different sensor applications

Section 2 shows the theoretical formulation of the MAFVRO algorithm for both the
proportional and non-proportional viscous damping models. The requirement of the algorithm is to
provide the system displacement, velocity and acceleration response matrices due to free vibration.
Wang and Cheng [20] illustrated the use of displacement sensor for the proportional viscous
damping model of MAFVRO. This section will show the implement for selecting different sensors,
in particular for the accelerometer that is frequently used.

8
 If the displacement sensor is used to measure the system displacement response, xr (t k ) = xr ,k ,

as illustrated in Fig. 1, the velocity and acceleration can be determined by finite difference method.
Table 1 shows the formulas to evaluate the velocity and acceleration. Wang and Cheng [20] showed
the matrix operations on response matrices, which are omitted here for brevity. The high order
formula can provide more accurate results as desired.

If the accelerometer is used, the acceleration at the r-th DOF xr (t k ) = xr ,k can be measured.

Table 2 shows the numerical formulas to evaluate the velocity and displacement, respectively.
Therefore, both Eq. (14) and Eq. (28) for the proportional and non-proportional viscous damping
models of the MAFVRO algorithms, respectively, can be obtained from the free vibration response
and solved for modal parameters. For example, if Simpson’s rule is adopted to obtain the velocity
and displacement from the acceleration as shown in Table 2, the minimum start number is k = 5
for the displacement response matrix shown in Eq. (11) because two previous data points are
required for each numerical integration operation.

4. Implementation of MAFVRO

This section introduces the development of MATLAB program to implement the MAFVRO
algorithm. Fig. 2(a) shows the solution flow chart for the MAFVRO application program. The steps
in the program are summarized as follows:
(1) Define the program parameters for the MAFVRO algorithm. Set up the start number of time

data k and total number of time data N k to be processed as revealed in Fig. 1.

(2) Select the analysis modes. Either the experiment or simulation can be selected. For the
experiment, the measured data, either the displacement or acceleration response, due to the
free vibration should be provided. For simulation, the free vibration simulation procedure is
shown in Fig. 2(b). To perform the response simulation, define the system matrices first, as
well as the initial conditions. Then, choose the time interval ∆t , i.e. the sampling frequency
is as follows:

1
fs = . (29)
∆t

Next, define the noise ratio (NR), which is the ratio of noise and the maximum response
amplitude for emulating the measured response containing noise as follows:

xr (t k ) = MAX ( xr (t k ) ) ‧NR‧RAN+ xr (t k ) . (30)

9
 where RAN is the normally distributed random number between -1 and 1; MAX ( xr (t k ) ) is

the maximum displacement in simulation. Finally, the response model for either the
proportional or non-proportional viscous damping can be specified and solved for both the
theoretical modal analysis and transient response analysis. The displacement response can
then be obtained and used for the MAFVRO algorithm applications. The simulated
acceleration can also be obtained such as by the Simpson’s rule as shown in Table 2.
(3) Define the type of sensor. Typical sensors including the displacement sensor and
accelerometer can be specified.
(4) Obtain the system free vibration response matrices, including the displacement, velocity and
acceleration as shown in Eqs. (11)-(13). Different numerical formula can be chosen for those
differential and integration methods as shown in Tables 1 and 2.
(5) Start MAFVRO main program. To obtain the modal parameters from the free vibration
response by choosing either the proportional or non-proportional viscous damping model of
the MAFVRO algorithms as revealed in Fig. 2(c). It is noted here that only the the r-th

natural frequency ( ω̂ r ) and its corresponding mode shape ( φ̂ r ) can be obtained for the

proportional viscous damping model, because the normal mode analysis is assumed, i.e. φ̂ r

is the real mode. For the non-proportional viscous damping model, the complex mode
analysis is adopted, so the equivalent natural frequency ( ωr ) and modal damping ratio ( ζ r )

can be obtained from Eqs. (23) and (24). The mode shape vector ( φ r ) is essentially

complex.
(6) Results comparison. The predicted modal parameters from the MAFVRO algorithm can be
compared with those from the theoretical modal analysis (TMA) or the conventional
experimental modal analysis (EMA). The prediction error for each modal frequency is
defined as follows:

fˆr − f r
εr = × 100% . (31)
fr

The damping ratio prediction error is also defined similarly. In particular, the modal
assurance criterion (MAC) value between two vectors is used to evaluate the prediction
effectiveness for mode shapes and given as follows:
2
φˆ Τrφ s
MAC (φˆ r , φ s ) = Τ * Τ * , r = 1,2,, n , s = 1,2,  , n . (32)
(φˆˆrφ r )(φ s φ s )

If MAC (φˆ r , φ s ) =1, the two mode shape vectors are perfectly proportional. For

10
 MAC (φˆ r , φ s ) =0, the two mode shape vectors are truly orthogonal.

5. Results and Discussions

This section will employ the developed MAFVRO algorithms for both the proportional and
non-proportional viscous damping models to obtain the structural modal parameters via simulation
data. Sections 5.1 and 5.2 show the applications of MAFVRO to the MDOF systems and the beam
structure, respectively.

5.1. MDOF system application

Fig. 3 shows the diagram of a n-DOF system model. Table 3 summarizes the system
parameters and the damping matrix C has the similar form to the stiffness matrix K, where mi =1
kg and k i =1,000,000 N/m. For the proportional viscous damping system model, the constants
defined in Eq. (4) are assumed as α =0.00001 and β =0.00001. Table 4 shows natural frequencies
and damping ratios determined by normal mode analysis for systems with different numbers of
DOFs. For the non-proportional viscous damping system model, only c1 =20(N.s/m) and else ci =0.
Table 5 reveals natural frequencies and damping ratios, as shown in Eqns. (23) and (24),
respectively, determined by complex mode analysis for different DOFs systems. One can observe
that there is a slight difference for natural frequencies between the proportional and
non-proportional viscous damping models, while the damping ratios are quite different for different
damping models and their assumptions. The following case studies of different DOFs systems for
both the proportional and non-proportional viscous damping system models are adopted
accordingly.

For the adoption of MAFVRO algorithms to obtain modal parameters via either the
proportional or non-proportional damping method, the system response matrices, including
displacement, velocity and acceleration, must be formulated first. In this paper unless noted, the
second order central formula is adopted to evaluate the velocity and acceleration for using the
displacement sensors to measure the response. For the use of accelerometers, the Simpson’s rule is
adopted to numerically obtain the velocity and displacement response from the acceleration.

Wang and Cheng [20] showed the system parameter effects of MAFVRO algorithm, including

the starting data point k and total number of time data N k as well as the sampling frequency f s ,

for using the displacement sensor by the proportional viscous damping MAFVRO method. Wang
and Cheng [20] showed the prediction of natural frequencies are exactly the same for different N k
and k , respectively, because the formulation of system response matrices as revealed in Eqns.

11
(11)-(13) are nearly exact solution. Therefore, different N k and k can be flexibly chosen to
formulate system matrices and result in good predictions of natural frequencies and mode shapes.
This work will mainly show the prediction by the non-proportional viscous damping MAFVRO
method and compare the performance between both models. The commonly used accelerometers in
experimental measurements for the MAFVRO application are shown additionally in this work.

Table 6 shows the prediction of modal parameters for the 10 DOFs system with the
non-proportional viscous damping effect by using displacement sensor and adopting the
non-proportional MAFVRO method. Table 6(a) reveals the effect of different number of data points
( N k ) for starting data point k=5 on the modal parameter prediction. Both predicted natural
frequency errors ε i and damping ratio errors are quite small and reveal the same for N k >100,
while the MACs for the predicted mode shapes in comparison to the theoretical ones are equal to
one for each mode, i.e. the perfect prediction for mode shapes. Table 6 (b) is the results for different
starting data points ( k ), when N k =100. The prediction of modal parameters is also very good. The
merit of MAFVRO method is that if the system response matrices can be formulated as accurately
as possible, the prediction of modal parameters will be almost exact solutions. The selection of
starting data point ( k ) and the number of data points ( N k ) can be flexibly chosen and result in very
good prediction of modal parameters by the MAFVRO method.

Table 7 shows the minimum sampling frequency for different numerical methods via different
sensor applications by the non-proportional viscous damping model of the MAFVRO algorithm.
One can see that the natural frequency predictions are within 2% errors for the second central
formula in the displacement sensor application, when f s is 2200 Hz, i.e. the frequency ratio
between the sampling frequency and the highest modal frequency is about 7.12. For the
accelerometer application, the use of Simpson’s rule can obtain accurate predictions if f s =3800 Hz,
i.e. the frequency ratio about 12.23. The higher order of numerical methods used in obtaining the
response matrices will result in better accuracy and require the smaller sampling frequency. Table 7
suggests the required sampling frequencies as the guideline for practical applications in using
displacement sensors and accelerometers.

Table 8 compares the prediction results from both the proportional and non-proportional
models of MAFVRO algorithms. The predicted natural frequencies are very good within 2% errors.
The damping ratio predictions are also good and can only be obtained by the non-proportional
model. It is noted that the mode shape prediction is also good. The MAC matrix between the
predicted and theoretical mode shape vectors revealed perfectly the unity matrix. The MAC value
for each mode is shown to be 1 and indicates the good prediction for mode shapes.

Table 9 shows the simulation results for considering the measurement noise effects in using
displacement sensors for different combinations of MAFVRO models and response simulation
models. The maximum predicted natural frequency errors are limited to about 2%, and the
maximum tolerant NR is shown for each case study. For the 10-DOF system, the maximum
12
tolerable NR is about 3% for the non-proportional MAFVRO model, when the non-proportional
response model is adopted. The predictions for damping ratios and mode shapes, which are not
shown for brevity, are generally good and similar to those results shown in Table 8 for the natural
frequency errors within 2%.

Table 10 reveals the maximum tolerable NR values in MAFVRO applications for using
different sensors by the non-proportional MAFVRO model. In general, the MAFVRO algorithm
can accommodate higher NR values for fewer DOFs system. For example, the use of displacement
sensors can tolerate up to 11% NR for the 3-DOF system, while there is only 3% NR for the
10-DOF system. NR values are much smaller for the accelerometer application due to twice of
numerical operations in simulations. One is the numerical differential operation on the displacement
data to obtain the velocity and acceleration responses, and another is the numerical integration
operation on the acceleration data to obtain the velocity and displacement in the MAFVRO
algorithm. If the theoretical response of x , x and x is used for simulating the use of
accelerometers to avoid the accumulated numerical errors, the accommodated NR (%) values will
be the same for both displacement sensors and accelerometers, since the formulated system
response matrices are exactly the same. In that case, the results in Table 10(b) will become the same
as Table 10(a). Here, Table 10(b) is shown to reveal the effect of the accumulated numerical errors.

In summary, the use of the developed MAFVRO algorithms in the MDOF system application
is feasible. The non-proportional viscous damping model for the MAFVRO algorithm developed in
this work is also shown to obtain the natural frequencies, damping ratios and mode shape vectors,
simultaneously, from the free vibration response only without the prior knowledge of system
matrices.

5.2. Beam structure application

A steel cantilever beam is considered. Fig. 4 shows the illustration of the beam structure and
Table 11 reveals its physical properties. The proportional viscous damping model for the beam is
adopted as formulated by Wang [24] that showed the derivation of modal analysis and transient
response analysis. The beam lateral displacement response due to the initial conditions is simulated
and used as the measured data to be applied to the developed MAFVRO algorithms for both the
proportional and non-proportional models. It is noted that the lateral displacement response can be
obtained with the exact solutions for free vibration analysis under the thin beam assumptions. The
velocity and acceleration of the beam are determined by the second order central formula as shown
in Table 1.

By applying the developed MAFVRO algorithms to the beam structure, the free vibration
response of the beam should be measured at a number of points over the beam. In this study, there
are 14 points along the beam length, i.e. m=14. Therefore, the number of DOFs becomes n=m.
13
Tables 12(a) and 12 (b) show the prediction results via the proportional model of MAFVRO
algorithm by using displacement sensors and accelerometers, respectively. Both the natural
frequency and mode shape predictions are very good for the first five modes. For the use of
displacement sensors Table 12(a) reveals the natural frequency prediction errors are less than 1%,
while Table 12(b) for the use of accelerometers results in about 2% error. The slightly higher errors
in using accelerometers are the result of accumulated numerical errors from the differentiation and
integration process in simulating the beam transient responses to formulate the system response
matrices for the MAFVRO application. The MAC matrices between the predicted and theoretical
mode shapes are nearly close to the unity matrices, i.e. the mode shapes show good comparisons
and possess the orthogonal properties. The modal parameters predictions are very good.

For the modal parameter prediction results from the non-proportional MAFVRO model as
shown in Tables 13(a) and 13(b) by using displacement sensors and accelerometers, respectively,
both natural frequencies and mode shapes also reveal very good predictions, while the damping
ratios may reveal high errors but in the reasonable range. Note that the damping errors in Table 13(b)
for the accelerometer application are larger than those in Table 13(a) for the displacement sensor is
due to the numerical operation in obtaining the simulated acceleration response as discussed in
Table 10 for MDOF systems. The presented MAFVRO algorithm shows an effective way to obtain
the modal parameters from the free vibration response only without the prior knowledge of system
matrices. Feeny and Liang [25] expanded the POD method [23] to the discrete and continuous
systems in random excitation; however, the requirement for a prior knowledge of the mass matrix is
its limitation. The MAFVRO algorithms for both proportional and non-proportional viscous
damping models are simple and straight forward as revealed in this work. In particular, the
MAFVRO methods require no prior knowledge of system matrices and only the free vibration
transient response in determining the structural modal parameters.

6. Conclusions

This paper extends the proportional viscous damping model of the MAFVRO algorithm [20] to
the non-proportional model. Additionally, the formulation for the use of accelerometers other than
the displacement sensors is provided. The MAFVRO algorithm development is theoretically
complete. The case studies for MDOF systems and the beam structure are demonstrated by
simulation results. Results show that the modal data for both the discrete and continuous systems
can be well identified by the MAFVRO algorithms. The developed modal analysis methods from
the free vibration response only are promising and have the potential for practical applications. For
the proportional damping model, the normal mode analysis is assumed and thus only the natural
frequencies and mode shapes can be obtained. The predicted mode shape vector is real. For the
non-proportional damping model, the complex mode analysis is adopted. The modal damping ratios
can also be obtained in addition to the natural frequencies and mode shapes. In particular, the mode

14
shape vector is complex and more appropriate for practical structures. This work enhances the
modal analysis technique by using the free vibration response only and shows the feasibility of the
MAFVRO in practical applications for both the discrete and continuous systems.

7. Acknowledgements

The authors are grateful for the financial support of this work under the contract number: NSC
97-2221-E-020-007 from National Science Council, Taiwan. The authors would like to thank two
anonymous reviewers and the Editor for their comments.

8. References

[1] D. J. Ewins, Modal testing: theory and practice, Second Edition, Research Studies Press Ltd.,
Letchworth, Hertfordshire, England, 2000.
[2] K. G. McConnell, Vibration testing theory and practice, John Wiley & Sons, Inc., New York,
1995.
[3] B. T. Wang, Structural modal testing with the use of various forms of actuators and sensors,
Mechanical Systems and Signal Processing, 12 (1998) 627–639.
[4] R. Brincker, L.M. Zhang, P. Andersen, Modal identification from ambient responses using
frequency domain decomposition. In: Proceedings of 18th international modal analysis
conference (IMAC), San Antonio, TX, USA, 2000. p. 625–30.
[5] R. Brincker, C. E. Ventura, P. Andersen, Damping estimation by frequency domain
decomposition. In: Proceedings of 19th international modal analysis conference (IMAC),
Orlando, FL, USA, 2001. p. 441–46.
[6] B. Cauberghe, P. Guillaume, P. Verboven, E. Parloo, Identification of modal parameters
including unmeasured forces and transient effects, Journal of Sound and Vibration 265 (2003)
609–625.
[7] F. Magalhaes, A. Cunha, E. Caetano, R. Brincker, Damping estimation using free decays and
ambient vibration tests, Mechanical Systems and Signal Processing (2009),
doi:10.1016/j.ymssp.2009.02.011.
[8] C. Gentile, N. Gallino, Ambient vibration testing and structural evaluation of an historic
suspension footbridge, Advances in Engineering Software 39 (2008) 356–366.
[9] S. M. Moore, J.C.S. Lai, K. Shankar, ARMAX modal parameter identification in the presence
of unmeasured excitation—II: Numerical and experimental verification, Mechanical Systems
and Signal Processing 21 (2007) 1616–1641.
[10] N. Larbi, J. Lardies, Experimental modal analysis of a structure excited by a random force,
Mechanical Systems and Signal Processing (2000) 14(2), 181-192.
[11] V. Papakos, S. D. Fassois, Multichannel identification of aircraft skeleton structures under

15
 unobservable excitation: a vector AR/ARMA framework, Mechanical Systems and Signal
Processing (2003) 17(6), 1271–1290.
[12] C. Devriendt, P. Guillaume, The use of transmissibility measurements in output-only modal
analysis, Mechanical Systems and Signal Processing 21 (2007) 2689–2696.
[13] C. Devriendt, P. Guillaume, Identification of modal parameters from transmissibility
measurements, Journal of Sound and Vibration 314 (2008) 343–356.
[14] E. Parloo, P. Verboven, P. Guillaume, M. Van Overmeire, Sensitivity-based
mass-normalization of mode shape estimates from output-only data, Proceedings of the
International Conference on Structural System Identification, Kassel, Germany, 2001, pp.
627–636.
[15] E. Parloo, P. Verboven, P. Guillaume, M. Van Overmeire, Force identification by means of
in-operation modal models, Journal of Sound and Vibration 262 (2003) 161–173.
[16] M. Abdelghani, M. Goursat, T. Biolchini, On-line modal monitoring of aircraft structures
under unknown excitation, Mechanical Systems and Signal Processing, 13(6) 1999 839-853.
[17] S. Naylor, J. E. Cooper, J. R. Wright, Modal parameter estimation of non-proportionally
damped systems using force appropriation, Proceedings of the IMAC-XV, International Modal
Analysis Conference, Orlando, FL. (1997)
[18] C. S. Huang, W. C. Su, Identification of modal parameters of a time invariant linear system by
continuous wavelet transformation, Mechanical Systems and Signal Processing 21 (2007)
1642–1664.
[19] J. Lardies, Modal parameter identification from output-only measurements, Mechanics
Research Communication 24(5) (1997) 521-528.
[20] B. T. Wang, D. K. Cheng, Modal analysis of MDOF system by using free vibration response
data only, Journal of Sound and Vibration, 311(3-5) (2008) 737-755.
[21] W. Zhou, D. Chelidze, A new method for vibration modal analysis, In: Proceeding of the SEM
Annual Conference and Exposition on Experimental and Applied Mechanics, Portland, OR,
USA, 2005, pp. 390–391.
[22] B. F. Feeny, R. Kappagantu, On the physical interpretation of proper orthogonal modes in
vibration, Journal of Sound and Vibration 211 (1998) 607–616.
[23] S. Han, B. Feeny, Application of proper orthogonal decomposition to structural vibration
analysis, Mechanical System and Signal Processing 17 (2003) 989–1001.
[24] B. T. Wang, Vibration analysis of a continuous system subject to generic forms of actuation
forces and sensing devices, Journal of Sound and Vibration, 319 (2009) 1222-1251.
[25] B. F. Feeny, Y. Liang, Interpreting proper orthogonal modes of randomly excited vibration
system, Journal of Sound and Vibration, 265 (2003) 953-966.

16